THE SUMMARIZING OF CLINICAL EXPERIMENTS BY SIGNIFICANCE LEVELS

For a controlled clinical experiment in which two alternative treatments are compared, the statistical report often culminates in a significance test of the null hypothesis of no difference between the treatments, and significance at the 5 per cent level is taken as positive evidence of difference. It is argued that such an experiment serves primarily an inferential purpose; it is not a simple decision procedure, although its effect on practice may be considered in relation to ethical issues. Statistical inference should not be identified with testing this null hypothesis, despite the emphasis on such tests by R. A. Fisher in his work on design of experiments. This null hypothesis often has no interest or credibility. Copyright © 1990 John Wiley & Sons, Ltd.